GPT 6 Sol and Luna arrive as builders ask for more room to work
OpenAI released GPT 6 Sol and Luna on September 22, 2026. Compare their published API costs, availability, and the usage question BridgeMind raised.
This article reports verified release details and BridgeMind’s own observations. Follow the references for the original posts and announcements.
What did OpenAI release?
OpenAI released GPT 6 Sol and GPT 6 Luna on September 22, 2026. Sol targets complex coding and agent workflows at a lower cost than Astra. Luna targets focused work at high volume. OpenAI says both are available in ChatGPT Work and Codex for eligible Plus, Pro, Business, Enterprise, and Edu plans, and in the API. Its announcement says they are not yet available in Chat at publication time.
For standard API requests with shorter context, OpenAI lists Sol at $2 per million input tokens and $10 per million output tokens. Luna is listed at $0.10 input and $0.50 output. OpenAI describes Sol and Luna as 50% cheaper than the promotional GPT 5.6 prices for their respective tiers. Requests with long context, cache writes, and other processing options have different rates, so check the current pricing page before estimating a production bill.
Why does this matter to coding agents?
On September 16, BridgeMind argued that GPT 6 Astra and Claude Fable 5.1 were capable but difficult to use continuously within its subscriptions. That post describes BridgeMind’s own workload and accounts, not a universal limit for every builder. Sol and Luna answer the broader need for model choices at different cost and capability levels; their release does not by itself prove how much longer a specific Codex subscription session will run.
OpenAI’s model catalog recommends Astra for the hardest tasks from start to finish, Sol for a balance of intelligence and cost, and Luna for volume where cost matters most. That suggests a practical routing experiment: keep a demanding bug investigation on Astra, try Sol for ordinary feature work, and use Luna for short edits with clear instructions. Review the output and allowance consumed before making the routing permanent.
BridgeMind’s September 16 post on the need for more usable coding session time.
View the original post on X
How should builders compare them?
Use the same repository task and acceptance checks for each model. Record the plan or API tier, reasoning effort, elapsed time, total tokens, cost or allowance consumed, and the amount of human rework. Run more than one task: a single attractive screenshot or one benchmark score cannot capture reliability across a codebase.
BridgeMind’s early X ranking placed Opus 5.5 and Astra above Sol for its preferences. Treat that as an initial opinion, not a controlled benchmark. The release question is whether Sol or Luna gives a builder enough quality at a price and usage level that keeps the work moving.
BridgeMind’s first ranking of Opus 5.5, Astra, Fable 5.1, and Sol on release day.
View the original post on X